Automatic detection and removal of personal identifiable information (PII) in arabic text datasets / Rania Chaib

Public ISBD Titre : Automatic detection and removal of personal identifiable information (PII) in arabic text datasets Type de document : document multimédia Auteurs : Rania Chaib, Auteur ; Mohamed El Habib Maicha, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 78 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Artificial intelligence and data science Langues : Anglais (eng) Mots-clés : Personally identifiable information (PII)  Text De-identification  Arabic named entity recognition (NER)  AraBERT  Hybrid deep learning  Data privacy  Sequence labeling Résumé : The automatic detection and anonymization of Personally Identifiable Information (PII) in Arabic text remains a challenging task due to the linguistic complexity of the Arabic language and the lack of annotated datasets. This thesis addresses this problem by proposing an end-to-end hybrid pipeline for Arabic PII extraction and de-identification. To overcome the limitation of domain-specific resources, a custom corpus of 5,000 sentences was constructed and annotated across eight PII categories. We systematically evaluated state-of-the-art transformer-based models, including AraBERT, CAMeLBERT, and hybrid neural models combining AraBERT with BiLSTM and CRF layers. To further enhance robustness in detecting structured entities, a post-processing module based on regular expressions (Regex) was integrated into the neural pipeline. Experimental results show that the AraBERT-based hybrid pipeline achieves the best performance, obtaining a Macro F1-score of 0.90 on the custom dataset and outperforming both CAMeLBERT and the hybrid neural architectures (AraBERT-BiLSTM and AraBERT-CRF). Finally, multiple de-identification strategies were implemented, including masking, random replacement, and a context-aware synthetic generation pipeline, enabling the transformation of sensitive Arabic text into anonymized form while preserving its usability. note de thèses : Mémoire de master en informatique Automatic detection and removal of personal identifiable information (PII) in arabic text datasets [document multimédia] / Rania Chaib, Auteur ; Mohamed El Habib Maicha, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 78 p. + 1 disque optique numérique (CD-ROM). Option : Artificial intelligence and data science Langues : Anglais (eng) Mots-clés : Personally identifiable information (PII)  Text De-identification  Arabic named entity recognition (NER)  AraBERT  Hybrid deep learning  Data privacy  Sequence labeling Résumé : The automatic detection and anonymization of Personally Identifiable Information (PII) in Arabic text remains a challenging task due to the linguistic complexity of the Arabic language and the lack of annotated datasets. This thesis addresses this problem by proposing an end-to-end hybrid pipeline for Arabic PII extraction and de-identification. To overcome the limitation of domain-specific resources, a custom corpus of 5,000 sentences was constructed and annotated across eight PII categories. We systematically evaluated state-of-the-art transformer-based models, including AraBERT, CAMeLBERT, and hybrid neural models combining AraBERT with BiLSTM and CRF layers. To further enhance robustness in detecting structured entities, a post-processing module based on regular expressions (Regex) was integrated into the neural pipeline. Experimental results show that the AraBERT-based hybrid pipeline achieves the best performance, obtaining a Macro F1-score of 0.90 on the custom dataset and outperforming both CAMeLBERT and the hybrid neural architectures (AraBERT-BiLSTM and AraBERT-CRF). Finally, multiple de-identification strategies were implemented, including masking, random replacement, and a context-aware synthetic generation pipeline, enabling the transformation of sensitive Arabic text into anonymized form while preserving its usability. note de thèses : Mémoire de master en informatique

---

1 (1 - 1 / 1) Par page : 25 50 100 200